Task processing method, text processing method, automatic question answering method, task processing model training method, information processing method based on task processing model and cloud training platform

Abstract

The embodiment of the invention provides a task processing method, a text processing method, an automatic question answering method, a task processing model training method, an information processing method based on a task processing model and a cloud training platform, which are applied to the technical field of computers, and the task processing method comprises the following steps: obtaining task data of a target task; the task data are input into a task processing model, a task processing result of the target task is obtained, the task processing model is obtained through training on the basis of multiple pieces of target sample data and target sample results of the multiple pieces of target sample data, and the target sample data are obtained through screening on the basis of repeated content detection results of the multiple pieces of sample data; the target sample result content is not repeated. The target sample data is obtained by screening based on the repeated content detection result, and the content of the target sample result is not repeated, so that the repeated illusion of the model is relieved on the premise of not damaging the processing capability of the model and not depending on external knowledge, and the accuracy and integrity of the task processing result are improved.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and in particular to a task processing method, a text processing method, an automatic question-answering method, a task processing model training method, an information processing method based on a task processing model, and a cloud training platform. Background Art

[0002] With the advancement of computer technology, large models have begun to shine. Their remarkable capabilities in language understanding, generation, interaction, and reasoning have led to their widespread application in natural language processing fields such as dialogue, translation, and code generation. However, large models also face the challenge of hallucinations. Hallucinations are a phenomenon that occurs in large models, particularly in natural language processing tasks such as text generation and question-answering. While the output of a model may appear reasonable and fluent, it may actually contain erroneous information, fabricated facts, or be significantly inconsistent with real-world conditions. This inconsistency between model output and real-world knowledge or user input is called hallucinations.

[0003] Currently, to address the problem of poor task processing accuracy caused by large model hallucinations, retrieval enhancement methods are often used to leverage external knowledge sources, allowing large models to rely on reliable knowledge sources for task processing. However, retrieval enhancement methods rely too heavily on external knowledge sources, and the hallucination problem inherent in large models remains unresolved. As a result, the accuracy of large models remains low, leading to poor task processing accuracy when directly using large models. Therefore, a highly accurate task processing solution is urgently needed. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a task processing method. One or more embodiments of this specification also relate to a text processing method, an automatic question-answering method, a task processing model training method, an information processing method based on a task processing model, a cloud training platform, a task processing device, a text processing device, an automatic question-answering device, a task processing model training device, an information processing device based on a task processing model, a computing device, a computer-readable storage medium, and a computer program product to address technical deficiencies in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a task processing method is provided, including:

[0006] Get the task data of the target task;

[0007] The task data is input into the task processing model to obtain the task processing result of the target task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0008] According to a second aspect of the embodiments of this specification, a text processing method is provided, including:

[0009] Get the text to be processed for the target text task;

[0010] The text to be processed is input into the task processing model to obtain the text processing result of the target text task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the content of the target sample results is not repeated.

[0011] According to a third aspect of the embodiments of this specification, there is provided an automatic question-answering method, comprising:

[0012] Get the pending questions for the target question-answering task;

[0013] The question to be processed is input into the task processing model to obtain the answer result of the target question-answering task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, and the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0014] According to a fourth aspect of the embodiments of this specification, a task processing model training method is provided, including:

[0015] Acquire multiple target sample data, wherein the target sample data is obtained by screening based on duplicate content detection results of the multiple sample data;

[0016] Inputting a plurality of target sample data into an initial processing model to obtain target sample prediction results of the plurality of target sample data;

[0017] The initial processing model is trained according to the target sample prediction result and the target sample results of multiple target sample data to obtain a trained task processing model, wherein the target sample result content is not repeated.

[0018] According to a fifth aspect of the embodiments of this specification, there is provided an information processing method based on a task processing model, which is applied to a cloud training platform, including:

[0019] Receiving a task generation request sent by a terminal device, wherein the task generation request includes request information;

[0020] Based on the request information, a task processing model is obtained, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is obtained by screening duplicate content detection results of the multiple sample data, and the target sample results have no duplicate content;

[0021] Based on the task processing model, task information is generated, wherein the task information is used by the terminal device to execute the target task.

[0022] According to a sixth aspect of the embodiments of this specification, there is provided a task processing device, including:

[0023] A first acquisition module is configured to acquire task data of a target task;

[0024] The first input module is configured to input task data into a task processing model to obtain a task processing result of a target task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on duplicate content detection results of multiple sample data, and the target sample result content is not repeated.

[0025] According to a seventh aspect of the embodiments of this specification, there is provided a text processing apparatus, comprising:

[0026] A second acquisition module is configured to acquire the to-be-processed text of the target text task;

[0027] The second input module is configured to input the text to be processed into the task processing model to obtain the text processing result of the target text task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0028] According to an eighth aspect of the embodiments of this specification, an automatic question-answering device is provided, comprising:

[0029] A third acquisition module is configured to obtain pending questions for the target question-answering task;

[0030] The third input module is configured to input the question to be processed into the task processing model to obtain the answer result of the target question-answering task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0031] According to a ninth aspect of the embodiments of this specification, a task processing model training device is provided, comprising:

[0032] A fourth acquisition module is configured to acquire a plurality of target sample data, wherein the target sample data is obtained by screening based on duplicate content detection results of the plurality of sample data;

[0033] A fourth input module is configured to input a plurality of target sample data into the initial processing model to obtain target sample prediction results of the plurality of target sample data;

[0034] The first training module is configured to train the initial processing model according to the target sample prediction result and the target sample results of multiple target sample data to obtain a trained task processing model, wherein the target sample result content is not repeated.

[0035] According to a tenth aspect of the embodiments of this specification, there is provided an information processing device based on a task processing model, which is applied to a cloud training platform, comprising:

[0036] A first receiving module is configured to receive a task generation request sent by a terminal device, wherein the task generation request includes request information;

[0037] a fifth acquisition module configured to acquire a task processing model based on the request information, wherein the task processing model is trained based on the plurality of target sample data and target sample results of the plurality of target sample data, the target sample data is obtained by screening duplicate content detection results of the plurality of sample data, and the target sample results have no duplicate content;

[0038] The first generating module is configured to generate task information based on the task processing model, wherein the task information is used for the terminal device to execute the target task.

[0039] According to an eleventh aspect of the embodiments of this specification, a cloud training platform is provided, including:

[0040] A request interface, configured to receive a task generation request sent by a terminal device, wherein the task generation request includes request information;

[0041] The response unit is used to obtain a task processing model based on the request information, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is screened based on the repeated content detection results of the multiple sample data, and the target sample result content is not repeated; based on the task processing model, task information is generated, wherein the task information is used for the terminal device to execute the target task.

[0042] According to a twelfth aspect of the embodiments of this specification, a computing device is provided, including:

[0043] memory and processor;

[0044] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method provided in the first aspect, the second aspect, the third aspect, the fourth aspect, or the fifth aspect are implemented.

[0045] According to the thirteenth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, which, when executed by a processor, implements the steps of the method provided in the first aspect, second aspect, third aspect, fourth aspect, or fifth aspect above.

[0046] According to the fourteenth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method provided in the first aspect, second aspect, third aspect, fourth aspect, or fifth aspect above.

[0047] One embodiment of the present specification provides a task processing method, comprising: obtaining task data for a target task; inputting the task data into a task processing model to obtain a task processing result for the target task, wherein the task processing model is trained based on multiple target sample data and target sample results for the multiple target sample data, the target sample data is screened based on duplicate content detection results for the multiple sample data, and the target sample results have no duplicate content. Because the target sample data for training the task processing model is screened based on duplicate content detection results for the multiple sample data, and the target sample results have no duplicate content, the task processing model effectively mitigates the duplication illusion problem, improves the precision of the task processing model, and further improves the accuracy and completeness of the task processing results, without compromising the task processing model's own processing capabilities and without relying on external knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is an architecture diagram of a task processing system provided by one embodiment of this specification;

[0049] Figure 2 This is an architecture diagram of another task processing system provided by one embodiment of this specification;

[0050] Figure 3 This is a flowchart of a task processing method provided by one embodiment of this specification;

[0051] Figure 4 This is a flowchart of duplicate content detection in a task processing method provided in one embodiment of this specification;

[0052] Figure 5 This is a process flow chart of a task processing method provided by one embodiment of this specification;

[0053] Figure 6 is a flowchart of a text processing method provided by one embodiment of this specification;

[0054] Figure 7 This is a flow chart of an automatic question-answering method provided by one embodiment of this specification;

[0055] Figure 8 This is a flowchart of a task processing model training method provided by one embodiment of this specification;

[0056] Figure 9 This is a flowchart of an information processing method based on a task processing model provided by one embodiment of this specification;

[0057] Figure 10 This is a schematic diagram of the structure of a cloud training platform provided by one embodiment of this specification;

[0058] Figure 11 This is a structural diagram of a task processing device provided by one embodiment of this specification;

[0059] Figure 12 This is a structural diagram of a text processing device provided by an embodiment of this specification;

[0060] Figure 13 This is a schematic diagram of the structure of an automatic question-answering device provided by one embodiment of this specification;

[0061] Figure 14 This is a structural diagram of a task processing model training device provided by one embodiment of this specification;

[0062] Figure 15 This is a schematic diagram of the structure of an information processing device based on a task processing model provided by an embodiment of this specification;

[0063] Figure 16 This is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0064] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0065] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "the," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0066] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0067] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0068] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. A large model can also be called a foundation model / foundation model (Foundation Model 1). The large model is pre-trained on a large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks and has good generalization capabilities, such as a large-scale language model (LLM, Large Language Model 1) and a multi-modal pre-training model (multi-modal pre-training model 1).

[0069] In actual applications, large models only require a small number of samples to fine-tune the pre-trained model and can be applied to different tasks. Large models can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, they can be applied to computer vision tasks such as visual question answering (VQA), image description (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0070] First, the terms involved in one or more embodiments of this specification are explained.

[0071] Knowledge compilation: refers to updating the knowledge representation within the model through fine-tuning or specific training methods to make it reflect the latest facts or correct the model's erroneous cognition.

[0072] Supervised Fine-Tunning (SFT) is a large model training technique that usually refers to fine-tuning a pre-trained model using labeled data on a specific task or target dataset to adapt it to a new, specific learning task.

[0073] Oracle model: In the field of natural language processing, an oracle model usually refers to an idealized model or system that has very good performance or capabilities in a certain aspect and can be used to evaluate the performance of other models or systems, such as the chatGPT (chat Generative Pre-trained Transformer) model.

[0074] Convolutional Neural Networks (CNN) model: a multi-layer deep learning model with forward propagation and backpropagation, and a convolution kernel (filter) for processing feature data.

[0075] Recurrent Neural Network (RNN) model: A recursive deep learning model that processes vector representations recursively and connects intermediate layers in a chain-like manner.

[0076] Long Short-Term Memory (LSTM) model: A deep learning model that has the ability to memorize both long-term and short-term information and has a convolutional filter for processing feature data.

[0077] Deep Self-Attention Model (Transformer Model): A deep learning architecture based on the attention mechanism for processing sequence data such as natural language.

[0078] Bidirectional Encoder Representations from Transformers (BERT) is a special Transformer model trained using a bidirectional Transformer encoder and large-scale unlabeled text data.

[0079] Repetition Penalty: This refers to a constraint imposed on the model during the generation process. It reduces the probability of generating consecutive repeated content by modifying the model's generation probability distribution. Specifically, if the model considers a word that has already appeared in a nearby position when generating the next word, the model's generation probability of that word will be reduced by the penalty factor. For example, if the repetition penalty parameter is set to a value greater than 1, the model will suppress repeated content; if it is set to less than 1, it may allow more repetition.

[0080] Data cleaning refers to a series of preprocessing operations performed on raw data during data analysis or data mining. Its purpose is to improve data quality and ensure the accuracy and reliability of subsequent analysis results. This process includes identifying and correcting (or deleting) erroneous, incomplete, inconsistent, duplicate, or irrelevant data in the dataset.

[0081] With the advancement of natural language processing technology, large models have achieved significant breakthroughs in text understanding and generation. However, they also face challenges such as hallucinations. Large model hallucinations include situations where the model-generated content is inconsistent with the real world or user instructions, or where the model repeatedly repeats previous output content. The danger of hallucinations is that seemingly realistic or reasonable answers generated by the model may actually be incorrect, confusing, or completely fabricated. These hallucinations make it impossible to guarantee the accuracy and completeness of the content generated by large models, which poses significant challenges to their practical application. Therefore, how to effectively detect and mitigate large model hallucinations has become a hot topic in current natural language research.

[0082] Currently, the repetition hallucination problem of large models is often mitigated through retrieval-enhanced generation and adjusting inference hyperparameters. However, retrieval-enhanced generation methods utilize external knowledge sources, allowing the model to generate answers based on reliable knowledge sources. This reliance on external capabilities persists, and the hallucination problem inherent in the large model remains, failing to truly alleviate it. Adjusting inference hyperparameters, particularly increasing the repetition penalty parameter during inference, forces the model to sample fewer or no previously output text tokens, thereby alleviating repetition hallucinations. However, this method relies on the repetition penalty parameter value set by external user input, and similarly fails to truly alleviate the hallucination problem inherent in the large model, while also damaging the model's normal capabilities.

[0083] In practical applications, hallucination output can be broadly categorized into three types: factual hallucination, faithful hallucination, and repetition hallucination. Factual hallucination refers to the model claiming that a false fact is true, or generating statements that have no basis in reality. Faithful hallucination refers to the model failing to follow the user's input instructions to complete the task, and the output deviates from the user's requirements. Repetition hallucination refers to the model repeatedly repeating itself after outputting a certain number of words when answering questions. See Table 1 below for an example of model hallucinations:

[0084] Table 1 Model hallucination examples

[0085]

[0086]

[0087] The following is a detailed analysis based on the examples in Table 1:

[0088] As shown in the example of factual hallucination, the model mistakenly believes that Edison invented the lightbulb. In fact, Edison improved the design of the lightbulb and was not the only inventor. The model has learned common human errors in the world corpus. Collective misconceptions are reflected in online posts, forums, and articles. As the model learns from this data in large quantities, it inevitably inherits and imitates these "collective hallucinations," leading to cognitive errors. These hallucinations primarily manifest as "inconsistency with facts" or "fabrication," resulting in discrepancies between the generated content and the real world. The model's factual hallucinations are largely due to inherent errors in the training data.

[0089] From the example of the faithful illusion, we can see that the user clearly indicated in the input that apples should be excluded, but the model still output "red apples" in the answer, reflecting the model's lack of ability to follow instructions and the model's excessive memory of repeated information in the training data. This illusion is mainly manifested in the existence of "inconsistency" in the generated content, that is, there is a deviation between the generated content and the user's input instructions and contextual content. The model directly remembers the data instead of understanding the data, thereby damaging its ability to understand the semantics of specific instructions.

[0090] As can be seen from the example of repetition hallucination, this phenomenon is particularly severe when using model-generated datasets to train small models. This is because the generated conversation data often contains expressions that do not conform to human language habits. For example, the model tends to list multiple concepts or nouns in its responses, or repeatedly repeat certain words. This type of data is difficult to learn and can easily confuse small models, leading to repetition hallucination, where the model repeatedly repeats the content it has listed after outputting a certain number of words.

[0091] The following rules can be seen from the three types of hallucinations of the model: the content of the model output largely reflects the distribution, regularity and tendency of the data; the model is very sensitive to errors and repeated information in the training data and is easily affected by negative data. Model creativity and model hallucination are actually only a fine line apart. In the phenomenon of model capability emergence, creativity and hallucination are essentially two-in-one concepts. Therefore, the embodiments of this specification focus on how to reduce the problem of model repetition hallucinations, and propose a knowledge editing scheme based on data cleaning. By constructing target sample data and target sample results of target sample data, and training a task processing model based on multiple target sample data and target sample results of multiple target sample data, it can effectively reduce the repetition hallucination problem of the task processing model without damaging the processing capability of the task processing model itself and without relying on external knowledge, improve the accuracy of the task processing model and the user experience of the task processing model in actual use, further improve the accuracy and completeness of the content generated by the task processing model, and is a more feasible and more practical task processing solution.

[0092] Specifically, when using the above-mentioned scheme for task processing, the task data of the target task can be obtained; the task data is input into the task processing model to obtain the task processing result of the target task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, and the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0093] In this specification, a task processing method is provided. This specification also involves a text processing method, an automatic question-answering method, a task processing model training method, an information processing method based on a task processing model, a cloud training platform, a task processing device, a text processing device, an automatic question-answering device, a task processing model training device, an information processing device based on a task processing model, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0094] See also Figure 1 , Figure 1 This is an architecture diagram of a task processing system provided by an embodiment of this specification. The task processing system may include a client 100 and a server 200;

[0095] The client 100 is used to send task data of the target task to the server 200;

[0096] The server 200 is configured to input the task data into a task processing model, obtain a task processing result for a target task, wherein the task processing model is trained based on a plurality of target sample data and target sample results of the plurality of target sample data, the target sample data is obtained by screening based on duplicate content detection results of the plurality of sample data, and the target sample results have no duplicate content; and send the task processing result to the client 100;

[0097] The client 100 is also used to receive the task processing result sent by the server 200.

[0098] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the task processing results are further improved.

[0099] See also Figure 2 , Figure 2 This is an architectural diagram of another task processing system provided in accordance with one embodiment of this specification. The task processing system may include multiple clients 100 and a server 200. The clients 100 may include terminal devices, and the server 200 may include cloud devices. Multiple clients 100 may establish communication connections via the server 200. In a task processing scenario, the server 200 provides task processing services between the multiple clients 100. The multiple clients 100 may act as either senders or receivers, communicating via the server 200.

[0100] Users can interact with the server 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100. In the task processing scenario, users can publish data streams to the server 200 through the client 100. The server 200 generates task processing results based on the data stream and pushes the task processing results to other clients with which communication has been established.

[0101] The client 100 and the server 200 are connected via a network. The network provides a medium for the communication link between the client 100 and the server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, or other processing before being released to the server 200.

[0102] The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5, Hypertext Markup Language 5) application, a light application (also known as a mini-program, a lightweight application), or a cloud application. The client 100 can be based on the software development kit (SDK) of the corresponding service provided by the server 200, such as developed based on the real-time communication (RTC) SDK. The client 100 can be deployed in an electronic device and needs to rely on the device to run or certain APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0103] The server 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers that provide background training to support models used on clients, and servers that process data sent by clients. It should be noted that the server 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0104] It is worth noting that the task processing result methods provided in the embodiments of this specification are generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server and thus execute the task processing result methods provided in the embodiments of this specification. In other embodiments, the task processing result methods provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0105] See also Figure 3 , Figure 3 This is a flowchart of a task processing method provided by an embodiment of this specification, which specifically includes the following steps:

[0106] Step 302: Obtain task data of the target task.

[0107] In one or more embodiments of this specification, during task processing, task data of a target task may be obtained, and then the task data may be processed using a task processing model to generate a task processing result of the target task.

[0108] Specifically, the target task can be tasks in different scenarios, such as intelligent question answering, abstract extraction, optical character recognition, object counting, and so on. Task data is the processing object of the task processing model. Task data can be data of different modalities, such as text data, image data, voice data, video data, and so on.

[0109] In practical applications, there are multiple ways to obtain task data for a target task, and the method to be used depends on the actual situation. This specification does not impose any restrictions on this method in the embodiments. In one possible implementation of this specification, the task data for the target task sent by the front-end user can be received. In another possible implementation of this specification, the task data for the target task can be read from other data acquisition devices or databases.

[0110] Step 304: Input the task data into the task processing model to obtain the task processing result of the target task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, and the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0111] In one or more embodiments of this specification, after obtaining the task data of the target task, the task data may be further input into a task processing model to obtain a task processing result of the target task.

[0112] Specifically, the task processing model refers to a deep learning model obtained by performing SFT on the initial processing model based on multiple target sample data and target sample results of the multiple target sample data. The task processing model may include large-scale model parameters, and therefore, the task processing model may be a large model. Task processing models include but are not limited to CNN models, RNN models, LSTM models, Transformer models, and BERT models. The task processing result is related to the target task. For example, if the target task is a summary extraction task, the task processing result is a summary of the long text; if the target task is a text translation task, the task processing result is the translated text.

[0113] It should be noted that the task processing model consists of an encoding unit and a decoding unit. When task data is input into the task processing model, the encoding unit first uses a self-attention mechanism to capture the complex dependencies between elements within the task data and gradually converts the task data into a high-level semantic embedding representation to obtain an encoding vector. Next, the decoding unit uses a self-attention and masking mechanism to gradually decode the encoding vector provided by the encoding unit to obtain the task processing result.

[0114] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the task processing results are further improved.

[0115] In an optional embodiment of the present specification, after inputting the task data into the task processing model and obtaining the task processing result of the target task, the following steps may be further included:

[0116] Mark the key information in the task processing results to obtain the updated task processing results;

[0117] Send the updated task processing results to the front-end user.

[0118] Specifically, key information is related to the modality of the task processing result. If the task processing result is a text modality, the key information is key text information such as keywords, key sentences, and keywords. If the task processing result is an image modality, the key information is key visual information such as visual key points (such as eyes, eyebrows, and corners of the mouth in the image). Taking the key information as key text information as an example, key text information refers to facts, data, concepts, sentences, or words that can highlight the theme of the text, convey the core idea, support the author's point of view, or describe the essence of the event. Key text information helps to quickly understand the main idea of ​​the text, grasp the structure of the article, answer related questions, and make effective decisions or extract information.

[0119] In practical applications, for example, when labeling key information in task processing results, the key text information can be bolded, highlighted, italicized, etc. For example, when labeling key information in task processing results, the key visual information can be marked with a background color, borders, etc.

[0120] By applying the solution of the embodiment of this specification, key information in the task processing result is marked to obtain an updated task processing result; the updated task processing result is sent to the front-end user so that the front-end user can view the result, thereby improving the user experience.

[0121] In an optional embodiment of the present specification, after inputting the task data into the task processing model and obtaining the task processing result of the target task, the following steps may be further included:

[0122] Send task processing results to front-end users;

[0123] Receive editing information sent by the front-end user, where the editing information is used to edit the task processing results;

[0124] The task processing result is edited according to the editing information to obtain the edited task processing result.

[0125] Specifically, editing information describes the front-end user's editing requirements for task processing results and can also be used to adjust parameters of the task processing model. Editing requirements include, but are not limited to, translation requirements, synonym replacement requirements, and interpretation requirements. Furthermore, editing information can also include editing information for key information within the task processing results.

[0126] For example, assuming that the task processing result is "You can buy your ticket on Saturday morning", and the editing information is "Please translate the result into English", the task processing result is translated according to the editing information to obtain the translated task processing result "You can buy your ticket on Saturday morning".

[0127] Furthermore, after obtaining the edited task processing result, the edited task processing result can be used to adjust parameters of the task processing model.

[0128] Using the solution of the embodiments of this specification, the task processing results are sent to the front-end user; editing information sent by the front-end user is received, wherein the editing information is used to edit the task processing results; and the task processing results are edited based on the editing information to obtain the edited task processing results. By editing the task processing results based on the editing information sent by the front-end user, human-computer interaction is enhanced, and the adaptability and flexibility of task processing are improved.

[0129] In an optional embodiment of the present specification, after inputting the task data into the task processing model and obtaining the task processing result of the target task, the following steps may be further included:

[0130] Send task processing results to front-end users;

[0131] Receive result feedback information sent by the front-end user, wherein the result feedback information is information that provides feedback on the task processing result based on the task information of the target task;

[0132] Based on the feedback information from the results, build a model to optimize the data;

[0133] Use model optimization data to adjust the parameters of the task processing model.

[0134] Specifically, result feedback information can include feedback on the content, quality, and completion of task processing results, reflecting the front-end user's true feelings and expectations about the task processing results. Result feedback information includes, but is not limited to, result quality evaluation information, corrected and accurate task processing results, and areas of model optimization. Model optimization data refers to accurately optimized sample data used to optimize the task processing model.

[0135] It should be noted that if the result feedback information is an accurate and corrected task processing result, the model optimization data can be constructed based on the task data of the target task and the corrected accurate task processing result. If the result feedback information is the optimization field of the model, such as the XXX field, the sample data of the XXX field can be obtained and the sample data of the XXX field can be determined as the model optimization data. Among them, the process of adjusting the parameters of the task processing model using the model optimization data is the same as the training process of the above-mentioned task processing model, and will not be repeated in this embodiment of the specification.

[0136] Using the solutions of the embodiments of this specification, task processing results are sent to front-end users; result feedback information from the front-end users is received, wherein the result feedback information is information providing feedback on the task processing results based on the task information of the target task; model optimization data is constructed based on the result feedback information; and the model optimization data is used to adjust the parameters of the task processing model. By collecting and utilizing the result feedback information, the performance of the task processing model is continuously optimized to more accurately meet the actual needs of the front-end users and improve the quality and accuracy of the final task processing results.

[0137] In an optional embodiment of the present specification, the above-mentioned construction of model optimization data based on the result feedback information may include the following steps:

[0138] Generate optimization prompt information based on the result feedback information, wherein the optimization prompt information is used to guide the front-end user to send model optimization data for optimizing the task processing model;

[0139] Send optimization prompt information to front-end users, and receive model optimization data sent by front-end users based on the optimization prompt information.

[0140] It should be noted that there are many ways to generate optimization prompt information, and the specific selection should be made according to the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation method of this specification, the pre-set optimization prompt information can be directly obtained, such as "I am very sorry to have brought you inaccurate information. Please point out the specific inaccuracies or provide the correct answers to relevant questions. I will correct and optimize my answers as soon as possible to better serve you." In another possible implementation method of this specification, the result feedback information can be type-identified to determine the information type of the result feedback information, and the information type can be further matched with the prompt type of each prompt information in the prompt information library, and the prompt information with the same prompt type as the information type can be determined as the optimization prompt information.

[0141] By applying the solution of the embodiments of this specification, optimization prompt information is generated based on the result feedback information; the optimization prompt information is sent to the front-end user, and model optimization data sent by the front-end user based on the optimization prompt information is received. The interactive guidance method for obtaining model optimization data improves interactivity with users and increases user satisfaction.

[0142] In an optional embodiment of this specification, a training method for a task processing model is described. That is, before inputting the task data into the task processing model and obtaining the task processing result of the target task, the following steps may be further included:

[0143] Acquire multiple target sample data;

[0144] Inputting a plurality of target sample data into an initial processing model to obtain target sample prediction results of the plurality of target sample data;

[0145] The initial processing model is trained according to the target sample prediction result and the target sample results of the plurality of target sample data to obtain a trained task processing model.

[0146] Specifically, the training method for the task processing model can be called knowledge editing. The target sample data carries the real data labels, which are the target sample results. Because the target sample results are non-repetitive, the target sample data can be considered pure data and serves as the processing target for the task processing model, guiding its training. Ideally, a task processing model trained using the target sample data will not experience repetition hallucinations.

[0147] It should be noted that the method for obtaining multiple target sample data may be to read a large amount of target sample data carrying target sample results from other data acquisition devices or databases. Alternatively, the method may be to receive a large amount of target sample data carrying target sample results input by a user. The method for obtaining multiple target sample data is selected based on actual circumstances and is not limited in any way in the embodiments of this specification.

[0148] In practical applications, when the initial processing model is trained based on the target sample prediction results and the target sample results of multiple target sample data, the loss value can be calculated based on the target sample prediction results and the multiple target sample data, and the model parameters of the initial processing model can be adjusted according to the loss value until the training process meets the preset stop conditions, and the trained task processing model is obtained. Among them, there are many functions for calculating the loss value, such as the cross entropy loss function, the L1 norm loss function, the maximum loss function, the mean square error loss function, the logarithmic loss function, etc., which are selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this. The preset stopping conditions include but are not limited to the loss value being less than or equal to the preset threshold and the number of iterations reaching the preset number of iterations, wherein the preset threshold and the preset number of iterations are selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0149] In one possible implementation of this specification, after calculating the loss value, the loss value is compared with a preset threshold. Specifically, if the loss value is greater than the preset threshold, it indicates that the difference between the target sample prediction result and the target sample result is large, and the initial processing model has poor predictive ability for the target sample data. In this case, the model parameters of the initial processing model can be adjusted, and the process returns to the step of inputting multiple target sample data into the initial processing model to obtain target sample prediction results for the multiple target sample data. The initial processing model is then trained until the loss value is less than or equal to the preset threshold, indicating that the difference between the target sample prediction result and the target sample result is small, and a preset stopping condition is met, thereby obtaining a trained task processing model.

[0150] In another possible implementation of this specification, in addition to comparing the loss value with a preset threshold, the number of iterations may also be used to determine whether the training of the current initial processing model is complete. Specifically, if the loss value is greater than the preset threshold, the model parameters of the initial processing model are adjusted, and the process returns to the step of inputting multiple target sample data into the initial processing model to obtain target sample prediction results for the multiple target sample data. Training of the initial processing model continues until the preset number of iterations is reached, at which point iterations are terminated to obtain a trained task processing model.

[0151] Using the solutions of the embodiments of this specification, an initial processing model is trained based on the target sample prediction results and the target sample results of multiple target sample data. If a preset stopping condition is not met, the initial processing model is trained continuously until the preset stopping condition is met, completing the training and obtaining a task processing model. By continuously adjusting the model parameters of the initial processing model, the resulting task processing model can be made more accurate.

[0152] In an optional embodiment of the present specification, since the parameters of the model will affect the frequency of the repetition hallucination problem, and repetition is more likely to occur when using low randomness parameters than when using high randomness parameters. Taking the sampling temperature as an example, when the sampling temperature is lowered, the model will be more inclined to generate tokens with the highest probability, which may cause the model to be too conservative and easy to fall into a loop or repeat the previously generated content, because the most likely sequence may sometimes be a repetition of an existing pattern. Therefore, the search results of the randomness parameters of the task processing model and the output stability of the model can be weighed to select better parameters for the model and further reduce the repetition hallucination problem of the model. That is, the above-mentioned initial processing model is trained based on the target sample prediction results and the target sample results of multiple target sample data. After obtaining the trained task processing model, the following steps can also be included:

[0153] Within a preset parameter range, a parameter search is performed on the task processing model, and the parameters of the task processing model are adjusted according to the parameter search results to obtain an adjusted task processing model.

[0154] Specifically, parameter search refers to an optimization method in the fields of machine learning and deep learning, whose goal is to find a set of optimal model parameters or hyperparameters that enable the model to achieve better performance on specific evaluation metrics. Parameter search is used to determine the optimal combination of model algorithm internal parameters (such as weights and biases in neural networks) and / or hyperparameters (such as learning rate, regularization strength, number of hidden layers, activation function, etc., which are variables set before training the model to control the model structure and training process).

[0155] It should be noted that, in the embodiments of this specification, the model randomness parameters (Top_P, Top_K, Temperature) can be adjusted and searched within the preset parameter range, wherein the Temperature parameter is used to adjust the randomness of the output. Increasing the setting of the Temperature parameter can make the generated results more random and innovative, while lowering the Temperature parameter can lead to more stable and repetitive results. The Top_K parameter limits the selection range when the model predicts the next word, limiting the model to select the predicted word from the most likely top K words. As the K value increases, the range of optional words becomes wider and the diversity of the results increases. Conversely, reducing the K value will reduce the range of optional words, making the generated results more inclined to words with a higher probability of appearing. The Top_P parameter limits the selection of the next word from a set of vocabulary sets when the probability accumulation reaches a given P value. A lower Top_P value makes the generated results more predictable and relevant, while a higher Top_P value increases the diversity and creativity of the results. The number of optional words in the above sampling method is dynamic, and the preset parameter range, K and P are set specifically according to actual conditions. The embodiments of this specification do not impose any restrictions on this.

[0156] In practical applications, methods for searching parameters for task processing models include but are not limited to: Grid Search: exhaustively enumerate all possible parameter combinations within the preset parameter range, train and verify the model performance one by one, and find a better combination. Random Search: randomly select points in the preset parameter range for sampling, which is more efficient than grid search. Bayesian Optimization: a method based on probability statistics, using a proxy model (such as a Gaussian process) to fit the relationship between parameters and model performance, thereby intelligently selecting the next parameter combination to be tested in the preset parameter range, aiming to minimize the number of experiments to find a better solution. Gradient-based Optimization: for those differentiable parameters, they can be updated by gradient descent or other optimization algorithms to achieve automatic adjustment of parameters.

[0157] By applying the solutions of the embodiments of this specification, a parameter search is performed on the task processing model within a preset parameter range, and the parameters of the task processing model are adjusted based on the parameter search results to obtain an adjusted task processing model. By weighing the search results of the randomness parameters of the task processing model and the output stability of the model, the optimal parameters of the model are selected, further reducing the problem of repeated hallucinations in the model.

[0158] In an optional embodiment of the present specification, after the initial processing model is trained based on the target sample prediction result and the target sample results of the plurality of target sample data to obtain the trained task processing model, the following steps may also be included:

[0159] Obtain multiple repeated evaluation sample data;

[0160] Inputting a plurality of repeated evaluation sample data into the task processing model to obtain a first evaluation result, and generating a first evaluation indicator of the task processing model based on the first evaluation result;

[0161] When the first evaluation indicator does not meet the model indicator condition, the process returns to the step of obtaining multiple target sample data until the first evaluation indicator meets the model indicator condition, thereby obtaining an optimized task processing model.

[0162] Specifically, duplicate evaluation sample data is used to quantify the duplication illusion of the task processing model. The duplicate evaluation sample results of duplicate evaluation sample data contain duplicate content, and therefore, the duplicate evaluation sample data can be understood as toxic data. Because duplicate evaluation sample data is toxic data, the model's processing of duplicate evaluation sample data is prone to duplication illusion. Therefore, the duplicate evaluation sample data is input into the task processing model. After obtaining a first evaluation result, the first evaluation result is tested for duplicate content to determine whether the duplication illusion problem of the current task processing model has been alleviated. The first evaluation result refers to the predicted result of the duplicate evaluation sample data. The first evaluation indicator is used to describe the content duplication of the first evaluation result, such as the duplication quality score and the first evaluation word frequency. The word frequency can be the number of times a word appears in the result, or it can be the relative frequency, that is, the value obtained by dividing the number of word appearances by the total number of words in the result. Model indicator conditions include, but are not limited to, the number of word appearances being less than a preset value and the word frequency being less than or equal to the preset frequency. The specific selection is based on actual circumstances and is not limited in the embodiments of this specification.

[0163] In practical applications, there are various ways to obtain multiple repeated evaluation sample data, and the specific method is selected based on the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, multiple repeated evaluation sample data can be received from a front-end user. In another possible implementation of this specification, multiple repeated evaluation sample data can be read from other data acquisition devices or databases.

[0164] Furthermore, after generating a first evaluation metric for the task processing model based on the first evaluation result, if the first evaluation metric satisfies the model metric condition, it indicates that the repetition hallucination problem of the current task processing model has been effectively alleviated. In this case, training can be terminated and the current task processing model can be used for task processing. If the first evaluation metric does not satisfy the model metric condition, it indicates that the repetition hallucination of the current task processing model still has a significant impact on model accuracy. In this case, the process can return to the step of obtaining multiple target sample data and train the current task processing model until the first evaluation metric satisfies the model metric condition, thereby obtaining an optimized task processing model.

[0165] Applying the solution of the embodiments of this specification, multiple repeated evaluation sample data are obtained; the multiple repeated evaluation sample data are input into the task processing model to obtain a first evaluation result, and based on the first evaluation result, a first evaluation index of the task processing model is generated; if the first evaluation index does not meet the model index condition, the step of obtaining multiple target sample data is returned until the first evaluation index meets the model index condition, thereby obtaining an optimized task processing model. By performing a repetition hallucination evaluation on the trained task processing model, the severity of the model's repetition hallucination can be effectively quantified, further ensuring the accuracy of the task processing model.

[0166] In an optional embodiment of the present specification, the obtaining of multiple target sample data may include the following steps:

[0167] Get multiple sample data;

[0168] Performing duplicate content detection on sample results of multiple sample data respectively to obtain duplicate content detection results;

[0169] According to the duplicate content detection result, a plurality of target sample data with non-duplicate sample result content are screened out from the plurality of sample data.

[0170] Specifically, duplicate content detection can be understood as duplicate hallucination detection, which is used to determine whether there is duplicate content (such as words, characters, and phrases) in the sample results. The duplicate content detection result is used to describe whether the sample results have duplicate content, such as whether the content is duplicate or not.

[0171] In practical applications, there are many ways to obtain multiple sample data, and the specific method is selected according to the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, multiple sample data can be received from a front-end user. In another possible implementation of this specification, multiple sample data can be read from other data acquisition devices or databases.

[0172] For example, an Oracle model can be used to collect a batch of model-generated data, which includes fields such as instruction, input, and output. The model-generated data is question-answer pair data that conforms to human conversation habits, and the model-generated data is determined as sample data.

[0173] Using the solution of the embodiments of this specification, multiple sample data are obtained; duplicate content detection is performed on the sample results of each of the multiple sample data to obtain duplicate content detection results; and based on the duplicate content detection results, multiple target sample data with non-duplicate sample results are filtered out from the multiple sample data. By filtering out the multiple target sample data with non-duplicate sample results from the sample data, the duplication illusion problem in the task processing model is alleviated.

[0174] In an optional embodiment of the present specification, performing duplicate content detection on sample results of the plurality of sample data to obtain duplicate content detection results may include the following steps:

[0175] Performing duplicate content detection on sample results of the plurality of sample data according to a word frequency screening strategy to obtain duplicate content detection results, wherein the word frequency screening strategy is used to perform duplicate content detection on the sample results according to the frequency of occurrence of words in the sample results; and / or,

[0176] The detection prompt information and the sample results of the plurality of sample data are input into the duplicate content detection model to obtain the duplicate content detection result.

[0177] Specifically, the duplicate content detection model can be a pre-trained large model or a deep learning model trained based on multiple training samples and the training results corresponding to the multiple training samples. Duplicate content detection models include but are not limited to CNN models, RNN models, LSTM models, Transformer models, and BERT models.

[0178] It should be noted that during duplicate content detection, word frequency filtering strategies can be used to examine sample data for output formats that do not conform to human conventions. These strategies primarily target common characteristics of duplicate data, such as identifying data with a high percentage of commas, a high total number of commas, or a high percentage of numbers as duplicate content. This strategy can be used to identify severely repetitive sample data.

[0179] Furthermore, because problems such as verbose language and missing information can lead to model hallucination, a duplicate content detection model can be used to perform quality scoring and filter out duplicate sample data. Specifically, by specifying low-quality data types and descriptions that can lead to model hallucination, such as verbose language, missing information, and irrelevant answers, in the detection prompt, this can prevent model hallucination caused by overfitting low-quality data during training.

[0180] In actual applications, it is observed that low-quality duplicate sample data often has three characteristics. First, it contains a lot of disgusting, negative or sensitive content; second, there is a serious semantic repetition; third, the writing is long-winded, using a large number of sentences to express the same meaning, and the information content is very low. Therefore, based on the above three characteristics, the embodiment of this specification designs a detection prompt message with item-by-item scoring, as shown below:

[0181] "As a strict data screener, please follow the following guidelines when scoring the quality of model training data. When scoring, please first determine whether there are the following errors. If any of the following situations occur, please set the score to 0 directly: 1. There are repugnant answers; 2. There are unhealthy contents; 3. There are sensitive topics. If there are no above errors, score each item according to the following scoring criteria: 1. Whether it is repeated (0 or 1 points): Is there any semantic repetition in the answer? If not, give 1 point, if so, give 0 points; 2. Whether it is long-winded (0 or 1 point): Based on the answer, judge whether the given answer is long-winded, with a large number of sentences expressing the same meaning. If not, give 1 point, if yes, give 0 points; score each item according to the above distribution and scoring criteria, and add up the total as the final score. Please be careful in scoring, make sure to strictly follow the scoring criteria, and avoid the situation where the wrong answer is not scored 0 points. Question [sample data]; Answer [sample result]; Score [duplicate content detection result]. Please use JSON format to score the answer, which contains two fields: score value (rate) and reason (reason): If you score 0, give the reason for the error, otherwise give the score and reason for each item. Scoring example: {"rate": score value, "reason": "If you score 0, give the reason for the error, otherwise give the score and reason for each item"}". Among them, JSON format is a lightweight data exchange format that is easy for humans to read and write, and easy for machines to parse and generate. It uses a completely language-independent text format to store and represent data, and can efficiently exchange data between different programming languages.

[0182] It is worth noting that the above detection prompt information has three major advantages: First, it can prevent inappropriate data from contaminating the model: by directly scoring answers containing inappropriate content as 0 points, this data can be prevented from contaminating the training set, which is crucial to ensuring the health of the model. Second, it simplifies the scoring system: the scoring criteria are binary (0 or 1 points), which reduces the random judgment of the model during annotation, improves the consistency of the scoring criteria and the efficiency of annotation. Third, the output is traceable and easy to parse: the scoring model needs to record the score and reasons in JSON format, which helps maintain the consistency of the record and facilitates subsequent review and analysis. Having the model use JSON format is also conducive to automated parsing.

[0183] By applying the solution of the embodiments of this specification, duplicate content detection is performed on sample results of multiple sample data according to a word frequency screening strategy to obtain duplicate content detection results; and / or, detection prompt information and sample results of multiple sample data are input into a duplicate content detection model to obtain duplicate content detection results. By detecting duplicate sample data that is likely to cause duplicate hallucinations in the sample data, and based on this method, the sample data is directionally separated into "toxic" duplicate sample data and "non-toxic" target sample data, and then a task processing model is trained using the target sample data, the duplicate hallucination problem of the task processing model is alleviated.

[0184] See also Figure 4 , Figure 4 This is a flowchart of duplicate content detection in a task processing method provided in one embodiment of this specification. The duplicate content detection process consists of two parts: word frequency screening strategy screening and model-based quality score screening. The execution process is as follows:

[0185] Frequency-based filtering: Based on sample data, a frequency-based filtering strategy is used. This strategy uses common characteristics of duplicate data to perform rule-based filtering. For example, data with a high percentage of commas, a high total number of commas, or a high percentage of numbers in the data are considered duplicate data. This frequency-based filtering strategy can roughly filter out some of the more serious duplicate sample data.

[0186] Model-based quality scoring screening: Obtain detection prompt information, filter the detection prompt information and the sample results of the remaining sample data based on the word frequency filtering strategy, input them into the duplicate content detection model for scoring, and filter out the duplicate sample data with lower scores.

[0187] It should be noted that by first filtering based on the word frequency strategy to obtain a rough screening result, and then filtering the remaining sample data based on the model's quality score, the data filtered out by the word frequency filtering strategy and the model's quality score is determined to be duplicate sample data. The unfiltered data is the target sample data, which will be used in the training process of the task processing model. Furthermore, the target sample data can be deduplicated to reduce errors and duplicate information in the target sample data.

[0188] In an optional embodiment of the present specification, after performing duplicate content detection on sample results of the plurality of sample data and obtaining duplicate content detection results, the following steps may be further included:

[0189] According to the duplicate content detection result, multiple duplicate sample data with duplicate sample result content are screened out from the multiple sample data;

[0190] The initial processing model is trained based on multiple repeated sample data to obtain a trained repeated processing model.

[0191] It should be noted that duplicate sample data refers to data containing repeated sample results. Because duplicate sample data contains duplicate sample results, it can be considered toxic data. Because duplicate evaluation sample data is toxic, the model's processing of duplicate sample data is prone to the phenomenon of "duplicate hallucination." In this case, training the initial processing model based on multiple duplicate sample data, and obtaining a trained duplicate processing model, further exacerbates the phenomenon of duplicate hallucination.

[0192] In actual applications, the implementation method of "training the initial processing model based on multiple repeated sample data to obtain a trained repeated processing model" is the same as the above-mentioned implementation method of "inputting multiple target sample data into the initial processing model to obtain target sample prediction results of multiple target sample data; training the initial processing model based on the target sample prediction results and the target sample results of multiple target sample data to obtain a trained task processing model", and will not be repeated in the embodiments of this specification.

[0193] Using the solution of the embodiments of this specification, based on the duplicate content detection results, multiple sample data with duplicate sample results are screened out from multiple sample data; based on the multiple duplicate sample data, an initial processing model is trained to obtain a trained duplicate processing model. By using toxic data to train a duplicate processing model with an exacerbated duplicate hallucination problem, the performance of the duplicate processing model can be compared with the performance of the task processing model to intuitively demonstrate the severity of the duplicate hallucination problem in the task processing model.

[0194] In an optional embodiment of the present specification, after filtering out multiple duplicate sample data with duplicate sample result content from multiple sample data according to the duplicate content detection result, the following steps may also be included:

[0195] Dividing the multiple repeated sample data to obtain multiple repeated evaluation sample data and multiple repeated training sample data;

[0196] Training the initial processing model based on multiple repeated sample data to obtain a trained repeated processing model may include the following steps:

[0197] The initial processing model is trained according to a plurality of repeated training sample data to obtain a trained repeated processing model.

[0198] It should be noted that in order to further quantify the severity of the repetition hallucination of the repeated processing model, the repeated sample data can be divided into multiple repeated evaluation sample data and multiple repeated training sample data, strictly ensuring that the repeated training sample data does not appear in the repeated evaluation sample data, making the hallucination quantification process more accurate.

[0199] In practical applications, there are many ways to divide multiple repeated sample data to obtain multiple repeated evaluation sample data and multiple repeated training sample data. The specific method is selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, a preset number of repeated sample data can be randomly screened from multiple repeated sample data as repeated evaluation sample data, and the unselected repeated sample data can be determined as repeated training sample data. In another possible implementation of this specification, a preset number of repeated sample data with serious content duplication can be used as repeated evaluation sample data, and the unselected repeated sample data can be determined as repeated training sample data.

[0200] By applying the solution of the embodiments of this specification, multiple repeated sample data are divided to obtain multiple repeated evaluation sample data and multiple repeated training sample data. The initial processing model is trained based on the multiple repeated training sample data to obtain a trained repeated processing model. By strictly ensuring that the repeated training sample data and the repeated evaluation sample data do not overlap, the hallucination quantification process is more accurate.

[0201] In an optional embodiment of the present specification, after the initial processing model is trained based on the plurality of repeated training sample data to obtain the trained repeated processing model, the following steps may also be included:

[0202] Inputting a plurality of repeated evaluation sample data into the repeated processing model to obtain a second evaluation result, and generating a second evaluation index of the repeated processing model based on the second evaluation result;

[0203] The first evaluation index and the second evaluation index of the task processing model are compared to obtain an index comparison result, wherein the index comparison result is used to describe the processing capability of the task processing model.

[0204] It should be noted that the implementation method of "inputting multiple repeated evaluation sample data into the repeated processing model to obtain a second evaluation result, and generating a second evaluation indicator of the repeated processing model based on the second evaluation result" is the same as the above-mentioned implementation method of "inputting multiple repeated evaluation sample data into the task processing model to obtain a first evaluation result, and generating a first evaluation indicator of the task processing model based on the first evaluation result", and this embodiment of the specification will not go into details about this.

[0205] In practical applications, after determining the first evaluation indicator of the task processing model and the second evaluation indicator of the repetition processing model, the first evaluation indicator and the second evaluation indicator of the repetition processing model can be compared, so as to determine the severity of the repetition illusion of the task processing model compared with the repetition processing model based on the indicator comparison results, thereby measuring the processing capability of the task processing model.

[0206] Furthermore, when comparing the first evaluation indicator and the second evaluation indicator, multiple repeated evaluation sample data can be additionally input into the initial processing model to obtain a third evaluation result, and based on the third evaluation result, a third evaluation indicator of the initial processing model is generated, and the first evaluation indicator, the second evaluation indicator and the third evaluation indicator are compared to obtain an indicator comparison result, so as to determine the severity of the repetition illusion of the task processing model compared with the initial processing model and the repeated processing model based on the indicator comparison result, thereby measuring and determining the processing capability of the task processing model.

[0207] For example, assuming that the first evaluation index of the task processing model is 68, the second evaluation index of the repetition processing model is 37.89, and the third evaluation index of the initial processing model is 60.86, this indicates that the task processing model's ability to cope with the repetition illusion problem can reach 68 points, the initial processing model's ability to cope with the repetition illusion problem can reach 60.86 points, and the repetition processing model's ability to cope with the repetition illusion problem can reach 37.89 points. Therefore, compared with the initial processing model, the repetition illusion problem of the repetition processing model is aggravated, while that of the task processing model is alleviated. Therefore, the repetition illusion problem of the task processing model is well alleviated, and the accuracy of the task processing model is improved.

[0208] Using the solution of the embodiments of this specification, multiple repeated evaluation sample data are input into the repetition processing model to obtain a second evaluation result. Based on the second evaluation result, a second evaluation index for the repetition processing model is generated. The first evaluation index and the second evaluation index of the task processing model are compared to obtain an index comparison result, where the index comparison result is used to describe the processing capability of the task processing model. By determining the severity of the repetition hallucination of the task processing model compared to the repetition processing model based on the index comparison result, the processing capability of the task processing model can be intuitively determined.

[0209] See also Figure 5 , Figure 5 This is a flowchart of a task processing method provided by one embodiment of this specification. The task processing method discusses in detail how to solve the model's repetition hallucination problem, focusing on aspects such as data source, training strategy, and model hyperparameter settings. Specifically, it includes:

[0210] First, duplicate content detection is used to separate duplicate sample data from the sample data, and duplicate target sample data is removed to reduce errors and duplicate information in the target sample data. To further quantify the severity of the model's duplication illusion, duplicate sample data is divided into duplicate evaluation sample data and duplicate training sample data, strictly ensuring that duplicate evaluation sample data does not appear in duplicate training sample data.

[0211] Secondly, the target sample data is used to edit the knowledge of the initial processing model, and the parameters in the initial processing model are fine-tuned through training to obtain a task processing model that effectively alleviates the problem of repetition hallucination. At the same time, the repeated training sample data is used to edit the knowledge of the initial processing model, and the parameters in the initial processing model are fine-tuned through training to obtain a repetition processing model that exacerbates the problem of repetition hallucination.

[0212] Then, the repetition hallucination of the task processing model and the repetition processing model is quantitatively calculated using the repeated evaluation sample data, thereby obtaining a first evaluation index of the task processing model and a second evaluation index of the repetition processing model. The first evaluation index and the second evaluation index are compared to determine the severity of the repetition hallucination of the task processing model and the repetition processing model.

[0213] Finally, in order to further reduce the repetition hallucination problem of the task processing model, the task processing model can be searched for parameters within the preset parameter range, and the parameters of the task processing model can be adjusted according to the parameter search results to obtain the adjusted task processing model.

[0214] By applying the solution of the embodiments of this specification, the toxic data in the sample data is separated to obtain pure target sample data, and then the target sample data is used in a supervised fine-tuning manner to correct the knowledge of the initial processing model, so that the model tends to output non-repetitive content, thereby alleviating the model's repetitive hallucination phenomenon. This can effectively reduce the model's repetitive hallucination without relying on external capabilities and without damaging the model's own dialogue ability.

[0215] The following combined Figure 6 , taking the application of the task processing method provided in this specification in the text processing scenario as an example, the task processing method is further explained. Figure 6 This is a flowchart of a text processing method provided by an embodiment of this specification, which specifically includes the following steps:

[0216] Step 602: Obtain the text to be processed of the target text task.

[0217] Step 604: Input the text to be processed into the task processing model to obtain the text processing result of the target text task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, and the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0218] It should be noted that the implementation of steps 602 to 604 is the same as the implementation of steps 302 to 304, and will not be described in detail in this embodiment of the specification.

[0219] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the repetition illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the text processing results are further improved.

[0220] In an optional embodiment of the present specification, after inputting the to-be-processed text into the task processing model and obtaining the text processing result of the target text task, the following steps may also be included:

[0221] Send text processing results to front-end users;

[0222] Receive result feedback information sent by the front-end user, wherein the result feedback information is information that provides feedback on the text processing result based on the task information of the target text task;

[0223] The result feedback information is sent to the cloud training platform, where the cloud training platform is used to adjust the parameters of the task processing model using the result feedback information.

[0224] In actual applications, after receiving the result feedback information sent by the front-end user, the result feedback information can be sent to the cloud training platform, and the cloud training platform can adjust the parameters of the task processing model based on the result feedback information. The method for the cloud training platform to adjust the parameters of the task processing model based on the result feedback information can refer to the above-mentioned implementation method of "constructing model optimization data based on the result feedback information; using the model optimization data to adjust the parameters of the task processing model", and will not be further described in this embodiment of the specification.

[0225] By applying the solution of the embodiments of this specification, the parameters of the task processing model are adjusted based on the result feedback information through the cloud training platform. This can reduce the system deployment and operation and maintenance costs while ensuring the accuracy and training efficiency of the task processing model, and provide users with convenient and efficient model training services.

[0226] The following combined Figure 7 , taking the application of the task processing method provided in this specification in the automatic question answering scenario as an example, the task processing method is further explained. Figure 7 This is a flowchart of an automatic question-answering method provided by an embodiment of this specification, which specifically includes the following steps:

[0227] Step 702: Obtain pending questions for the target question-answering task.

[0228] Step 704: Input the question to be processed into the task processing model to obtain the answer result of the target question-answering task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, and the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0229] It should be noted that the implementation of steps 702 to 704 is the same as the implementation of steps 302 to 304, and will not be described in detail in this embodiment of the specification.

[0230] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the response results are further improved.

[0231] In an optional embodiment of the present specification, before obtaining the pending questions for the target question-answering task, the following steps may be further included:

[0232] Sending question-and-answer prompt information to the front-end user, wherein the question-and-answer prompt information is used to guide the front-end user to send pending questions for processing the target question-and-answer task;

[0233] Receive pending questions for the target question-and-answer task sent by front-end users based on question-and-answer prompt information.

[0234] It's important to note that when automated Q&A begins, a Q&A prompt message can be sent to the front-end user. This prompt allows the front-end user to understand the automated Q&A process or related Q&A products, allowing them to enter questions that are more relevant to their needs. For example, the Q&A prompt message could read, "Hello, I'm your automated Q&A assistant. I'm ready to answer all your questions. Please enter your question directly."

[0235] Furthermore, after sending the question and answer prompt information to the front-end user, the pending questions of the target question and answer task sent by the front-end user based on the question and answer prompt information can be received, so that the pending questions are input into the task processing model to obtain the answer result of the target question and answer task.

[0236] Using the solution of the embodiments of this specification, a question-and-answer prompt is sent to a front-end user; and a pending question for a target question-and-answer task, sent by the front-end user based on the question-and-answer prompt, is received. The question-and-answer prompt guides the front-end user to enter the pending question to be solved, thereby enhancing human-computer interaction and ensuring that the pending question accurately reflects the user's actual needs.

[0237] See also Figure 8 , Figure 8 This is a flowchart of a task processing model training method provided by an embodiment of this specification, which specifically includes the following steps:

[0238] Step 802: Acquire multiple target sample data, wherein the target sample data is obtained by screening based on duplicate content detection results of multiple sample data.

[0239] Step 804: Input the multiple target sample data into the initial processing model to obtain target sample prediction results of the multiple target sample data.

[0240] Step 806: Train the initial processing model based on the target sample prediction result and the target sample results of multiple target sample data to obtain a trained task processing model, wherein the target sample result content is not repeated.

[0241] It should be noted that the implementation of steps 802 to 806 is the same as that of the above Figure 3The training method of the task processing model in the provided embodiments is the same, and will not be described in detail in the embodiments of this specification.

[0242] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated and the accuracy of the task processing model is improved.

[0243] See also Figure 9 , Figure 9 This is a flowchart of an information processing method based on a task processing model provided by an embodiment of this specification. The method is applied to a cloud training platform and specifically includes the following steps:

[0244] Step 902: Receive a task generation request sent by a terminal device, wherein the task generation request includes request information.

[0245] Specifically, the cloud training platform is an infrastructure based on cloud computing technology that provides large-scale data processing and high-performance computing resources for training, optimizing, and deploying various machine learning models, particularly deep learning models. On the cloud training platform, users can upload data, select or customize algorithm models, and efficiently train and verify models through distributed computing capabilities. The cloud training platform receives task generation requests from terminal devices, obtains the corresponding task processing model based on the request information, and generates task information based on the task processing model. The cloud training platform can quickly respond to the needs of different tasks, call the appropriate processing model for task processing, and ultimately generate high-quality task processing results.

[0246] A task generation request is a request instruction sent by a terminal device to the cloud training platform, requesting the cloud training platform to generate task information for the target task. A task generation request typically includes the data to be processed, the task type, the expected output format, and the request information. For example, when a user selects the "Text Translation" function on the cloud training platform's front-end interface and uploads text, the cloud training platform constructs a task generation request that includes information such as the text, the task type (i.e., text translation), and the language type for translation.

[0247] Request information refers to the parameters or descriptive information related to the target task carried in the task generation request. Request information is used to guide the cloud training platform to correctly identify and execute the requested task information. Request information includes, but is not limited to, the target task scenario identifier, task model identifier, or sample data for the target generation task.

[0248] Step 904: Based on the request information, obtain a task processing model, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is screened based on duplicate content detection results of multiple sample data, and the target sample result content is not repeated.

[0249] In an optional embodiment of this specification, the above-mentioned obtaining of the task processing model based on the request information may include the following steps:

[0250] Based on the task scenario identifier, a target scenario template is determined from a plurality of preset scenario templates, and based on the target scenario template, a task processing model is searched from a model library, wherein the model library stores a plurality of processing models, and the request information includes the task scenario identifier of the target task;

[0251] or,

[0252] Based on the task model identifier, the task processing model is searched from the model library, wherein the request information includes the task model identifier of the target task.

[0253] Specifically, the task scenario identifier refers to a unique or specific label used to distinguish different task application scenarios. In the embodiment of this specification, the task scenario identifier is part of the request information, and the cloud training platform can select a target scenario template that matches the request information from a series of preset scenario templates based on the task scenario identifier to generate task information. For example, if the task scenario identifier is "text translation", it means that the terminal device wants to translate the uploaded text, then a text translation scenario template can be selected from multiple preset scenario templates based on the task scenario identifier.

[0254] Preset scenario templates are predefined standard configuration scenario templates for different task application scenarios. Each template contains model information and task processing flow that matches the task application scenario. The cloud training platform stores a series of preset scenario templates to quickly respond to task generation requests for different scenarios. Different preset scenario templates correspond to different task types, model information, and processing flows, ensuring that the pre-training platform can automatically obtain the model and process configuration that best suits the current request information based on the task scenario identifier. For example, the preset scenario templates may include a template specifically for legal text processing, which contains the model information and processing flow of a trained legal model.

[0255] A target scenario template is a scenario template that matches the task scenario identifier. When parsing a task generation request, the cloud training platform can locate the corresponding target scenario template based on the task scenario identifier and select the corresponding task processing model and other related configuration information from the model library based on the model information included in the target scenario template. For example, if the task scenario identifier is "Financial Report Analysis," the target scenario template is a template that contains the model information and related configuration parameters of the financial report analysis model.

[0256] The model library is a centralized repository for deep learning models, which have been trained and optimized to solve various processing tasks. On the cloud training platform, the model library stores a large number of processing models, including but not limited to text classification models, text translation models, and text analysis models. Furthermore, the processing models in the model library can be divided into different versions based on their applicable scenarios. For example, the model library may contain multiple versions of text analysis models, such as text analysis models for financial data and text analysis models for social data.

[0257] A task model identifier is a unique or specific label used to distinguish models applicable to different tasks. For example, the task model identifier could be "financial data." Based on this identifier, you can search the model library for task processing models applicable to financial data.

[0258] For example, suppose a user selects the "text summary extraction" function through the front-end interface of an e-commerce application, and the application sends a task generation request to the cloud training platform. The request information contains the task scenario identifier "text summary extraction". After receiving this task generation request, the cloud training platform identifies the task scenario identifier as "text summary extraction". The cloud training platform finds a target scenario template that matches "text summary extraction" from multiple preset scenario templates. This target scenario template is pre-configured with model information and processing flow suitable for the text summary extraction task. Based on the target scenario template, the cloud training platform can obtain a pre-trained text summary extraction model from the model library and load relevant parameters and configuration files. Furthermore, the cloud training platform can generate task information based on the information in the target scenario template. The task information includes but is not limited to details such as the model address, input data processing method, and output result specifications, so that the terminal device can correctly call the text summary extraction model and perform the text summary extraction task.

[0259] Using the solutions of the embodiments of this specification, a target scenario template is determined from multiple preset scenario templates based on a task scenario identifier. Based on the target scenario template, a task processing model is searched from a model library, which stores multiple processing models. Alternatively, a task processing model is searched from the model library based on the task model identifier. By integrating cloud computing technology with predefined task scenario templates, task model identifiers, and model library resources, a flexible, efficient, and standardized task processing mechanism is achieved.

[0260] In another optional embodiment of the present specification, in addition to selecting a pre-trained task processing model from a model library, a task processing model adapted to the needs of the end user can be trained specifically based on sample data provided by the end user, that is, the request information includes sample data of the target task;

[0261] Based on the request information, obtaining the task processing model may include the following steps:

[0262] Based on the sample data, the initial processing model corresponding to the target task is trained to obtain a trained task processing model.

[0263] It should be noted that the implementation method of "training the initial processing model corresponding to the target task based on sample data to obtain a trained task processing model" is the same as the training method of the above-mentioned task processing model, and will not be repeated in this embodiment of the specification.

[0264] By applying the solution of the embodiment of this specification, the initial processing model corresponding to the target task is trained based on sample data to obtain a trained task processing model, thereby ensuring that the task processing model better meets user needs and ensures the accuracy of the task processing model.

[0265] Step 906: Generate task information based on the task processing model, wherein the task information is used for the terminal device to execute the target task.

[0266] Specifically, task information is generated by the cloud training platform after parsing the received task generation request. This information contains the model configuration and processing flow required to execute the target task. Based on this information, the terminal device or other server-side components can correctly use the task processing model to process the target task.

[0267] It should be noted that when generating task information based on a task processing model, the task processing model can be directly packaged to obtain the task information. Alternatively, the model information of the task processing model can be obtained and used to construct the task information based on the model information. This model information includes model parameter configuration, input data processing methods, expected output specifications, possible intermediate steps involved, and other auxiliary information.

[0268] For example, in a text summary extraction task, the task information may include the address of the selected text summary extraction model, the storage location of the input text, the target path of the output summary, and other parameters such as other environmental configurations required for the operation of the text summary extraction model. This information enables the terminal device to correctly load the text summary extraction model on a local or remote server and perform the text summary extraction task.

[0269] Using the solutions of the embodiments of this specification, a task generation request is received from a terminal device, wherein the task generation request includes request information; a task processing model is obtained based on the request information; and task information is generated based on the task processing model, wherein the task information is used for the terminal device to perform the target task. By using a cloud training platform that integrates cloud computing technology, task information for the terminal device to perform the target task is generated. This ensures the quality and efficiency of target task processing while reducing system deployment and operation and maintenance costs, providing users with convenient and efficient task information services.

[0270] See also Figure 10 , Figure 10 100 is a schematic diagram of the structure of a cloud training platform provided by an embodiment of this specification, the cloud training platform includes a request interface 1002 and a response unit 1004;

[0271] The request interface 1002 is used to receive a task generation request sent by a terminal device, wherein the task generation request includes request information;

[0272] Response unit 1004 is used to obtain a task processing model based on the request information, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is screened based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated; based on the task processing model, task information is generated, wherein the task information is used for the terminal device to execute the target task.

[0273] In an optional embodiment of the present specification, the cloud training platform further includes a model library, wherein the model library stores a plurality of processing models;

[0274] The response unit is specifically used to determine the target scene template from multiple preset scene templates based on the task scene identifier, and based on the target scene template, search for the task processing model from the model library, wherein the model library stores multiple processing models, and the request information includes the task scene identifier of the target task; or, based on the task model identifier, search for the task processing model from the model library, wherein the request information includes the task model identifier of the target task.

[0275] It should be noted that the implementation method of "the response unit determines the target scenario template from multiple preset scenario templates based on the task scenario identifier, and searches for the task processing model from the model library based on the target scenario template, wherein the model library stores multiple processing models, and the request information includes the task scenario identifier of the target task; or searches for the task processing model from the model library based on the task model identifier, wherein the request information includes the task model identifier of the target task" can refer to the above Figure 9 The implementation method of the provided embodiment of "based on the task scenario identifier, determining the target scenario template from multiple preset scenario templates, and based on the target scenario template, searching the task processing model from the model library, wherein the model library stores multiple processing models, and the request information includes the task scenario identifier of the target task; or, based on the task model identifier, searching the task processing model from the model library, wherein the request information includes the task model identifier of the target task" will not be repeated in the embodiments of this specification.

[0276] By applying the solution of the embodiments of this specification, the cloud training platform generates task information for terminal devices to perform target tasks, which can reduce system deployment and operation and maintenance costs while ensuring the quality and efficiency of target task processing, and provide users with convenient and efficient task information services.

[0277] Corresponding to the above-mentioned task processing method embodiment, this specification also provides a task processing device embodiment, Figure 11 This is a structural diagram of a task processing device provided by an embodiment of this specification. Figure 11 As shown, the device includes:

[0278] A first acquisition module 1102 is configured to acquire task data of a target task;

[0279] The first input module 1104 is configured to input task data into a task processing model to obtain a task processing result of a target task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on duplicate content detection results of multiple sample data, and the target sample result content is not repeated.

[0280] Optionally, the device also includes: a first sending module, configured to send the task processing results to a front-end user; receiving result feedback information sent by the front-end user, wherein the result feedback information is information that provides feedback on the task processing results based on the task information of the target task; constructing model optimization data based on the result feedback information; and using the model optimization data to adjust the parameters of the task processing model.

[0281] Optionally, the first sending module is further configured to generate optimization prompt information based on the result feedback information, wherein the optimization prompt information is used to guide the front-end user to send model optimization data for optimizing the task processing model; send the optimization prompt information to the front-end user, and receive the model optimization data sent by the front-end user based on the optimization prompt information.

[0282] Optionally, the device further includes: a marking module configured to mark key information in the task processing result to obtain an updated task processing result; and send the updated task processing result to a front-end user.

[0283] Optionally, the device also includes: a second sending module, configured to send the task processing result to the front-end user; receive editing information sent by the front-end user, wherein the editing information is used to edit the task processing result; edit the task processing result according to the editing information to obtain the edited task processing result.

[0284] Optionally, the device also includes: a second training module, configured to obtain multiple target sample data; input the multiple target sample data into the initial processing model to obtain target sample prediction results of the multiple target sample data; train the initial processing model based on the target sample prediction results and the target sample results of the multiple target sample data to obtain a trained task processing model.

[0285] Optionally, the device further includes: a search module configured to perform parameter search on the task processing model within a preset parameter range, and adjust the parameters of the task processing model according to the parameter search results to obtain the adjusted task processing model.

[0286] Optionally, the device also includes: a first evaluation module, configured to obtain multiple repeated evaluation sample data; input the multiple repeated evaluation sample data into the task processing model to obtain a first evaluation result, and generate a first evaluation indicator of the task processing model based on the first evaluation result; when the first evaluation indicator does not meet the model indicator condition, return to execute the step of obtaining multiple target sample data until the first evaluation indicator meets the model indicator condition, and obtain the optimized task processing model.

[0287] Optionally, the second training module is further configured to obtain multiple sample data; perform duplicate content detection on the sample results of the multiple sample data respectively to obtain duplicate content detection results; and filter out multiple target sample data whose sample result content is not duplicated from the multiple sample data based on the duplicate content detection results.

[0288] Optionally, the second training module is further configured to perform duplicate content detection on the sample results of multiple sample data according to a word frequency screening strategy to obtain duplicate content detection results, wherein the word frequency screening strategy is used to perform duplicate content detection on the sample results according to the frequency of occurrence of words in the sample results; and / or, input the detection prompt information and the sample results of multiple sample data into a duplicate content detection model to obtain duplicate content detection results.

[0289] Optionally, the device also includes: a third training module, configured to filter out multiple repeated sample data with repeated sample result content from multiple sample data based on the repeated content detection results; train the initial processing model based on the multiple repeated sample data to obtain a trained repeated processing model.

[0290] Optionally, the device also includes: a division module, configured to divide multiple repeated sample data to obtain multiple repeated evaluation sample data and multiple repeated training sample data; a third training module, further configured to train the initial processing model based on the multiple repeated training sample data to obtain a trained repeated processing model.

[0291] Optionally, the device also includes: a second evaluation module, configured to input multiple repeated evaluation sample data into the repeated processing model to obtain a second evaluation result, and generate a second evaluation indicator of the repeated processing model based on the second evaluation result; compare the first evaluation indicator and the second evaluation indicator of the task processing model to obtain an indicator comparison result, wherein the indicator comparison result is used to describe the processing capability of the task processing model.

[0292] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the task processing results are further improved.

[0293] The above is a schematic scheme of a task processing device of this embodiment. It should be noted that the technical scheme of the task processing device and the technical scheme of the task processing method described above are of the same concept. For details not described in detail in the technical scheme of the task processing device, please refer to the description of the technical scheme of the task processing method described above.

[0294] Corresponding to the above text processing method embodiment, this specification also provides a text processing device embodiment, Figure 12 This is a structural diagram of a text processing device provided by an embodiment of this specification. Figure 12As shown, the device includes:

[0295] The second acquisition module 1202 is configured to acquire the to-be-processed text of the target text task;

[0296] The second input module 1204 is configured to input the text to be processed into the task processing model to obtain the text processing result of the target text task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0297] Optionally, the device also includes: a third sending module, configured to send the text processing results to a front-end user; receive result feedback information sent by the front-end user, wherein the result feedback information is information that provides feedback on the text processing results based on task information of the target text task; and send the result feedback information to a cloud training platform, wherein the cloud training platform is used to adjust parameters of the task processing model using the result feedback information.

[0298] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the repetition illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the text processing results are further improved.

[0299] The above is a schematic diagram of a text processing device according to this embodiment. It should be noted that the technical solution of the text processing device and the technical solution of the above-mentioned text processing method are based on the same concept. For details not described in detail in the technical solution of the text processing device, please refer to the description of the technical solution of the above-mentioned text processing method.

[0300] Corresponding to the above-mentioned automatic question-answering method embodiment, this specification also provides an automatic question-answering device embodiment, Figure 13 This is a schematic diagram of the structure of an automatic question-answering device provided by an embodiment of this specification. Figure 13 As shown, the device includes:

[0301] The third acquisition module 1302 is configured to obtain pending questions for the target question-answering task;

[0302] The third input module 1304 is configured to input the question to be processed into the task processing model to obtain the answer result of the target question-answering task, wherein the task processing model is trained based on multiple target sample data and target sample results of multiple target sample data, the target sample data is obtained by screening based on the repeated content detection results of multiple sample data, and the target sample result content is not repeated.

[0303] Optionally, the device also includes: a fourth sending module, configured to send question and answer prompt information to the front-end user, wherein the question and answer prompt information is used to guide the front-end user to send pending questions for processing the target question and answer task; and receive pending questions of the target question and answer task sent by the front-end user based on the question and answer prompt information.

[0304] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated, the accuracy of the task processing model is improved, and the accuracy and completeness of the response results are further improved.

[0305] The above is a schematic diagram of an automatic question-answering device according to this embodiment. It should be noted that the technical solution of this automatic question-answering device and the technical solution of the automatic question-answering method described above are based on the same concept. For details not described in detail in the technical solution of the automatic question-answering device, please refer to the description of the technical solution of the automatic question-answering method described above.

[0306] Corresponding to the above-mentioned task processing model training method embodiment, this specification also provides a task processing model training device embodiment, Figure 14 This is a structural diagram of a task processing model training device provided by an embodiment of this specification. Figure 14 As shown, the device includes:

[0307] The fourth acquisition module 1402 is configured to acquire a plurality of target sample data, wherein the target sample data is obtained by screening based on duplicate content detection results of the plurality of sample data;

[0308] The fourth input module 1404 is configured to input a plurality of target sample data into the initial processing model to obtain target sample prediction results of the plurality of target sample data;

[0309] The first training module 1406 is configured to train the initial processing model according to the target sample prediction result and the target sample results of multiple target sample data to obtain a trained task processing model, wherein the target sample results are not repeated.

[0310] By applying the solution of the embodiments of this specification, since the target sample data for training the task processing model is obtained by screening the duplicate content detection results of multiple sample data, the target sample result content is not repeated. Therefore, without damaging the processing capability of the task processing model itself and without relying on external knowledge, the duplicate illusion problem of the task processing model is effectively alleviated and the accuracy of the task processing model is improved.

[0311] The above is a schematic diagram of a task processing model training device according to this embodiment. It should be noted that the technical solution of the task processing model training device and the technical solution of the task processing model training method described above are based on the same concept. For details not described in detail in the technical solution of the task processing model training device, please refer to the description of the technical solution of the task processing model training method described above.

[0312] Corresponding to the above method embodiment, this specification also provides an information processing device embodiment based on the task processing model, Figure 15 This is a structural diagram of an information processing device based on a task processing model provided by an embodiment of this specification. Figure 15 As shown, the device is applied to a cloud training platform and includes:

[0313] The first receiving module 1502 is configured to receive a task generation request sent by a terminal device, wherein the task generation request includes request information;

[0314] a fifth acquisition module 1504 configured to acquire a task processing model based on the request information, wherein the task processing model is trained based on the plurality of target sample data and target sample results of the plurality of target sample data, the target sample data is obtained by screening duplicate content detection results of the plurality of sample data, and the target sample results have no duplicate content;

[0315] The first generating module 1506 is configured to generate task information based on the task processing model, wherein the task information is used for the terminal device to perform the target task.

[0316] Optionally, the fifth acquisition module 1504 is further configured to determine a target scene template from multiple preset scene templates based on the task scenario identifier, and search for a task processing model from a model library based on the target scene template, wherein the model library stores multiple processing models and the request information includes the task scenario identifier of the target task; or, search for a task processing model from the model library based on the task model identifier, wherein the request information includes the task model identifier of the target task.

[0317] Optionally, the request information includes sample data of the target task; the fifth acquisition module 1504 is further configured to train the initial processing model corresponding to the target task based on the sample data to obtain a trained task processing model.

[0318] Using the solutions of the embodiments of this specification, a task generation request is received from a terminal device, wherein the task generation request includes request information; a task processing model is obtained based on the request information; and task information is generated based on the task processing model, wherein the task information is used for the terminal device to perform the target task. By using a cloud training platform that integrates cloud computing technology, task information for the terminal device to perform the target task is generated. This ensures the quality and efficiency of target task processing while reducing system deployment and operation and maintenance costs, providing users with convenient and efficient task information services.

[0319] The above is a schematic diagram of an information processing device based on a task processing model according to this embodiment. It should be noted that the technical solution of the information processing device based on the task processing model and the technical solution of the information processing method based on the task processing model are based on the same concept. For details not described in detail in the technical solution of the information processing device based on the task processing model, please refer to the description of the technical solution of the information processing method based on the task processing model.

[0320] Figure 16 16 is a block diagram of a computing device according to one embodiment of the present disclosure. Components of the computing device 1600 include, but are not limited to, a memory 1610 and a processor 1620. The processor 1620 is connected to the memory 1610 via a bus 1630, and a database 1650 is used to store data.

[0321] Computing device 1600 also includes an access device 1640 that enables computing device 1600 to communicate via one or more networks 1660. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1640 may include one or more of any type of network interface, wired or wireless (e.g., a network interface card (NIC)), such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a World Wide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0322] In one embodiment of the present specification, the above components of the computing device 1600 and Figure 16 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 16 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0323] Computing device 1600 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). Computing device 1600 may also be a mobile or stationary server.

[0324] Among them, processor 1620 is used to execute computer programs / instructions, which, when executed by the processor, implement the steps of the above-mentioned task processing method or text processing method or automatic question and answer method or task processing model training method or information processing method based on the task processing model.

[0325] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of this computing device and the technical schemes of the aforementioned task processing method, text processing method, automatic question-answering method, task processing model training method, and information processing method based on task processing model are based on the same concept. For details not described in detail in the technical scheme of the computing device, please refer to the description of the technical scheme of the aforementioned task processing method, text processing method, automatic question-answering method, task processing model training method, or information processing method based on task processing model.

[0326] An embodiment of the present specification also provides a computer-readable storage medium storing a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned task processing method, text processing method, automatic question-answering method, task processing model training method, or information processing method based on a task processing model.

[0327] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical schemes of the above-mentioned task processing method, text processing method, automatic question-answering method, task processing model training method, and information processing method based on the task processing model are of the same concept. For details not described in detail in the technical scheme of the storage medium, please refer to the description of the technical scheme of the above-mentioned task processing method, text processing method, automatic question-answering method, task processing model training method, or information processing method based on the task processing model.

[0328] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned task processing method or text processing method or automatic question-answering method or task processing model training method or information processing method based on the task processing model.

[0329] The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical scheme of this computer program product and the technical schemes of the aforementioned task processing method, text processing method, automatic question-answering method, task processing model training method, and information processing method based on task processing model are based on the same concept. For details not described in detail in the technical scheme of the computer program product, please refer to the description of the technical scheme of the aforementioned task processing method, text processing method, automatic question-answering method, task processing model training method, or information processing method based on task processing model.

[0330] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0331] The computer instructions include computer program codes, which may be in source code form, object code form, executable files, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0332] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0333] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0334] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A task processing method, comprising: Get the task data of the target task; The task data is input into a task processing model to obtain a task processing result of the target task, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is obtained by screening based on duplicate content detection results of multiple sample data, and the target sample result content is not duplicated.

2. The method according to claim 1, further comprising: Sending the task processing result to the front-end user; Receiving result feedback information sent by the front-end user, wherein the result feedback information is information providing feedback on the task processing result based on the task information of the target task; Building a model to optimize data based on the result feedback information; The model optimization data is used to adjust parameters of the task processing model.

3. The method according to claim 2, wherein constructing a model optimization data based on the result feedback information comprises: Generate optimization prompt information based on the result feedback information, wherein the optimization prompt information is used to guide the front-end user to send model optimization data for optimizing the task processing model; The optimization prompt information is sent to the front-end user, and the model optimization data sent by the front-end user based on the optimization prompt information is received.

4. The method according to claim 1, further comprising: Marking key information in the task processing result to obtain an updated task processing result; The updated task processing result is sent to the front-end user.

5. The method according to claim 1, after inputting the task data into a task processing model and obtaining the task processing result of the target task, further comprising: Sending the task processing result to the front-end user; receiving editing information sent by the front-end user, wherein the editing information is used to edit the task processing result; The task processing result is edited according to the editing information to obtain an edited task processing result.

6. The method according to claim 1, before inputting the task data into the task processing model to obtain the task processing result of the target task, further comprising: Acquire multiple target sample data; Inputting the plurality of target sample data into an initial processing model to obtain target sample prediction results of the plurality of target sample data; The initial processing model is trained according to the target sample prediction result and the target sample results of the multiple target sample data to obtain a trained task processing model.

7. The method according to claim 6, wherein after training the initial processing model based on the target sample prediction result and the target sample results of the plurality of target sample data to obtain a trained task processing model, the method further comprises: Within a preset parameter range, a parameter search is performed on the task processing model, and the parameters of the task processing model are adjusted according to the parameter search result to obtain an adjusted task processing model.

8. The method according to claim 6, wherein after training the initial processing model based on the target sample prediction result and the target sample results of the plurality of target sample data to obtain a trained task processing model, the method further comprises: Obtain multiple repeated evaluation sample data; Inputting the plurality of repeated evaluation sample data into the task processing model to obtain a first evaluation result, and generating a first evaluation index of the task processing model based on the first evaluation result; If the first evaluation indicator does not meet the model indicator condition, return to the step of obtaining multiple target sample data until the first evaluation indicator meets the model indicator condition, thereby obtaining an optimized task processing model.

9. The method according to claim 6, wherein obtaining a plurality of target sample data comprises: Get multiple sample data; Performing duplicate content detection on the sample results of the plurality of sample data respectively to obtain duplicate content detection results; According to the duplicate content detection result, a plurality of target sample data with non-duplicate sample result content are screened out from the plurality of sample data.

10. The method according to claim 9, wherein performing duplicate content detection on each of the sample results of the plurality of sample data to obtain duplicate content detection results comprises: performing duplicate content detection on the sample results of the plurality of sample data according to a word frequency screening strategy to obtain duplicate content detection results, wherein the word frequency screening strategy is used to perform duplicate content detection on the sample results according to the frequency of occurrence of words in the sample results; and / or, The detection prompt information and the sample results of the plurality of sample data are input into a duplicate content detection model to obtain the duplicate content detection result.

11. The method according to claim 9, further comprising: performing duplicate content detection on each of the sample results of the plurality of sample data, and obtaining duplicate content detection results; According to the duplicate content detection result, screening out a plurality of duplicate sample data having duplicate sample result content from the plurality of sample data; The initial processing model is trained according to the multiple repeated sample data to obtain a trained repeated processing model.

12. The method according to claim 11, after filtering out a plurality of duplicate sample data having duplicate sample result content from the plurality of sample data according to the duplicate content detection result, further comprising: Dividing the plurality of repeated sample data to obtain a plurality of repeated evaluation sample data and a plurality of repeated training sample data; The step of training the initial processing model based on the plurality of repeated sample data to obtain a trained repeated processing model includes: The initial processing model is trained according to the multiple repeated training sample data to obtain a trained repeated processing model.

13. The method according to claim 12, further comprising: training the initial processing model based on the plurality of repeated training sample data to obtain a trained repeated processing model; Inputting the plurality of repeated evaluation sample data into the repeated processing model to obtain a second evaluation result, and generating a second evaluation index of the repeated processing model according to the second evaluation result; The first evaluation indicator and the second evaluation indicator of the task processing model are compared to obtain an indicator comparison result, wherein the indicator comparison result is used to describe the processing capability of the task processing model.

14. A text processing method, comprising: Get the text to be processed for the target text task; The text to be processed is input into a task processing model to obtain a text processing result of the target text task, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is obtained by screening based on duplicate content detection results of multiple sample data, and the content of the target sample result is not repeated.

15. The method according to claim 14, after inputting the to-be-processed text into the task processing model and obtaining the text processing result of the target text task, further comprising: Sending the text processing result to the front-end user; Receiving result feedback information sent by the front-end user, wherein the result feedback information is information providing feedback on the text processing result based on the task information of the target text task; The result feedback information is sent to a cloud training platform, wherein the cloud training platform is used to adjust parameters of the task processing model using the result feedback information.

16. An automatic question-answering method, comprising: Get the pending questions for the target question-answering task; The question to be processed is input into a task processing model to obtain an answer result of the target question-answering task, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is obtained by screening based on duplicate content detection results of multiple sample data, and the target sample result content is not repeated.

17. The method according to claim 16, before obtaining the pending questions for the target question-answering task, further comprising: Sending question-and-answer prompt information to the front-end user, wherein the question-and-answer prompt information is used to guide the front-end user to send questions to be processed for processing the target question-and-answer task; Receive the pending questions of the target question-and-answer task sent by the front-end user based on the question-and-answer prompt information.

18. A task processing model training method, comprising: Acquire a plurality of target sample data, wherein the target sample data is obtained by screening based on duplicate content detection results of the plurality of sample data; Inputting the plurality of target sample data into an initial processing model to obtain target sample prediction results of the plurality of target sample data; The initial processing model is trained according to the target sample prediction result and the target sample results of the multiple target sample data to obtain a trained task processing model, wherein the target sample result content is not repeated.

19. An information processing method based on a task processing model, applied to a cloud training platform, comprising: Receiving a task generation request sent by a terminal device, wherein the task generation request includes request information; Based on the request information, a task processing model is obtained, wherein the task processing model is trained based on a plurality of target sample data and target sample results of the plurality of target sample data, the target sample data is obtained by screening based on duplicate content detection results of the plurality of sample data, and the target sample results have no duplicate content; Based on the task processing model, task information is generated, wherein the task information is used for the terminal device to perform the target task.

20. The method according to claim 19, wherein obtaining a task processing model based on the request information comprises: Based on the task scenario identifier, a target scenario template is determined from a plurality of preset scenario templates, and based on the target scenario template, a task processing model is searched from a model library, wherein the model library stores a plurality of processing models, and the request information includes the task scenario identifier of the target task; or, Based on the task model identifier, a task processing model is searched from the model library, wherein the request information includes the task model identifier of the target task.

21. The method according to claim 19, wherein the request information includes sample data of the target task; The acquiring of the task processing model based on the request information includes: Based on the sample data, an initial processing model corresponding to the target task is trained to obtain a trained task processing model.

22. A cloud training platform, comprising a request interface and a response unit; The request interface is used to receive a task generation request sent by a terminal device, wherein: The task generation request includes request information; The response unit is used to obtain a task processing model based on the request information, wherein the task processing model is trained based on multiple target sample data and target sample results of the multiple target sample data, the target sample data is screened based on duplicate content detection results of multiple sample data, and the target sample result content is non-duplicate; based on the task processing model, task information is generated, wherein the task information is used for the terminal device to execute the target task.

23. The cloud training platform according to claim 22, further comprising a model library, wherein: The model library stores a plurality of processing models; The response unit is specifically used to determine a target scene template from multiple preset scene templates based on the task scene identifier, and search for a task processing model from a model library based on the target scene template, wherein the model library stores multiple processing models, and the request information includes the task scene identifier of the target task; or, based on the task model identifier, search for a task processing model from the model library, wherein the request information includes the task model identifier of the target task.

24. A computing device comprising: memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 13 or any one of claims 14 to 15 or any one of claims 16 to 17 or claim 18 or any one of claims 19 to 21 are implemented.

25. A computer-readable storage medium storing a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any one of claims 1 to 13 or any one of claims 14 to 15 or any one of claims 16 to 17 or claim 18 or any one of claims 19 to 21.

26. A computer program product comprising a computer program / instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 13 or any one of claims 14 to 15 or any one of claims 16 to 17 or claim 18 or any one of claims 19 to 21.

Images